WO2022201451A1

WO2022201451A1 - Detection device and detection method

Info

Publication number: WO2022201451A1
Application number: PCT/JP2021/012632
Authority: WO
Inventors: 海斗笹尾
Original assignee: 株式会社日立国際電気
Priority date: 2021-03-25
Filing date: 2021-03-25
Publication date: 2022-09-29
Also published as: JPWO2022201451A1; JP7436752B2

Abstract

To provide a detection device that prevents normality/abnormality determination from being made with an object to be detected being overlooked even when the object to be detected is shown in an extremely small area of an image, a detection device 100 comprises an image acquisition unit 202 that acquires a first image including an object to be detected, an image processing unit 301 that outputs a second image obtained by processing the first image, an AE processing unit 303 that outputs a third image obtained by processing at least the second image out of the first image and the second image by using a trained autoencoder, a calculation processing unit 304 that calculates a difference between the first image and the third image and outputs a fourth image, and a determination unit 305 that makes normality/abnormality determination on the basis of the fourth image.

Description

Detection device and detection method

The present invention relates to a detection device and detection method using an autoencoder (AE).

Conventionally, technology related to anomaly detection that automatically judges whether a manufactured product is good or bad using an imaging device such as an industrial camera has been known. In recent years, anomaly detection technology using models using artificial intelligence (AI), especially deep learning, has become mainstream. In particular, in fields where it is difficult to collect a large amount of abnormal data, a technique called an autoencoder that learns only with non-defective data is used. Based on this technique, anomaly detection techniques using various techniques such as DAE (Denoising Auto Encoder) and VAE (Variational Auto Encoder) have been proposed. In these methods, only normal data is used for learning so that the output is equal to the input. During operation, the image data to be judged is input, and if it is restored normally, it is judged to be good. If not, it is determined to be defective (there is an abnormal part).

For example, Patent Document 1 discloses a detection device that improves detection accuracy by appropriately performing preprocessing and selection of learning data and model selection when detecting log anomalies using an autoencoder. The detection device includes a preprocessing section, a generation section, and a detection section. The preprocessing unit processes learning data and detection target data. The generation unit generates a normal state model by deep learning based on the learning data processed by the preprocessing unit. The detection unit calculates the degree of abnormality based on the output data obtained by inputting the data to be detected processed by the preprocessing unit into the model, and detects an abnormality in the data to be detected based on the degree of abnormality.

Japanese Patent Application Laid-Open No. 2020-140580

However, the causes of defects that are mixed in with manufactured products are often very small scratches, defects, and processing deviations in the images that show them. There is a problem that it is difficult to detect the cause indicated in such a small area even by using the above-described technique, and the cause is often overlooked.

Therefore, the present invention has been devised in view of such circumstances. It is to provide a detection device and a detection method that prevent the determination of

In order to solve the above problems, a detection device for detecting normality/abnormality of a detection target included in an image includes an image acquisition unit for acquiring a first image including the detection target, and a second image obtained by processing the first image. an image processing unit that outputs an image; an AE processing unit that outputs a third image obtained by processing at least a second image out of the first image and the second image by a learned autoencoder; the first image and the third image; and a calculation processing unit that outputs a fourth image, and a determination unit that determines normality/abnormality based on the fourth image.
According to this, by performing a determination based on a fourth image showing the difference between the first image to be determined and the third image processed by the autoencoder to the second image processed by the image, It is possible to provide a detection device that prevents the detection target from being overlooked and normality/abnormality determination even when the target to be detected is shown in an extremely small area.

Furthermore, the image processing section may be characterized by outputting a second image obtained by partially processing the first image.
According to this, since the second image is obtained by partially processing the first image, which is the determination target, the detection target that existed in the determination target appears in the third image. is highlighted and displayed, and it is possible to prevent such a detection target from being overlooked.

Further, the image processing section outputs a plurality of second images by performing a plurality of different processing processes on the first image, and the AE processing section processes the plurality of second images, and processes the plurality of second images. a plurality of third images corresponding to each of the above, and the calculation processing unit may be characterized by calculating differences between the first image and the plurality of third images and outputting one fourth image.
According to this, by synthesizing a plurality of different third images obtained by processing a plurality of differently processed second images with an autoencoder to generate one fourth image, and making a determination based on the fourth image , even an extremely small detection target can be displayed in an emphasized manner by superimposing the corresponding portion, and it is possible to prevent such a detection target from being overlooked.

Further, the calculation processing unit may be characterized by outputting one fourth image by calculating a logical product of a plurality of calculated differences.
According to this, one fourth image is generated by calculating the logical product of a plurality of differences, and a determination is made based on this fourth image, so that even an extremely small detection target is emphasized by overlapping the relevant part. It is possible to prevent such a detection target from being overlooked.

Furthermore, the image processing unit may be characterized by performing arbitrary two or more different processing from a set of brightness conversion processing, blurring processing, edge enhancement processing, and alpha blending processing.
According to this, by performing a plurality of different processing processes on the first image that is the determination target, the detection target that existed in the determination target appears in all the third images in common. By doing so, it is displayed in an emphasized manner, and it is possible to prevent such a detection target from being overlooked.

Furthermore, it may be characterized by further comprising an imaging device that captures an image of the detection target, and the image acquiring unit acquires the first image from the imaging device.
According to this, it is possible to provide a detection device integrated with an imaging device.

In order to solve the above problems, there is provided a detection method for detecting normality/abnormality of a detection target contained in an image, comprising: acquiring a first image containing the detection target; outputting; outputting a third image obtained by processing at least the second image among the first image and the second image by a learned autoencoder; calculating a difference between the first image and the third image; , outputting a fourth image, and making a normal/abnormal determination based on the fourth image.
According to this, by performing a determination based on a fourth image showing the difference between the first image to be determined and the third image processed by the autoencoder to the second image processed by the image, Even if an object to be detected is shown in an extremely small area, it is possible to provide a detection method that prevents such a detection object from being overlooked and normal/abnormal judgments to be made.

As described above, according to the present invention, even if an object to be detected is shown in an extremely small area in an image, it is possible to prevent such a detection object from being overlooked and to determine normality/abnormality. It is possible to provide a detection device and a detection method for detecting.

The figure which shows the usage condition of the abnormality detection system of 1st embodiment which concerns on this invention. 1 is a functional block diagram of an anomaly detection system according to a first embodiment of the present invention; FIG. FIG. 3 is a functional block diagram of an image processing unit of the detection device according to the first embodiment of the present invention; FIG. 4 is a diagram showing the relationship among an input image, a processed image, and an AE processing result image in the detection device of the first embodiment according to the present invention; FIG. 5 is a diagram showing the relationship among an input image, an AE processing result image, a difference result image, and a synthesis result image in the detection device of the first embodiment according to the present invention; 4 is a flowchart of processing in the calculation processing unit of the detection device according to the first embodiment of the present invention; 1 is a hardware configuration diagram of an anomaly detection system according to a first embodiment of the present invention; FIG. A diagram showing the functionality of a typical autoencoder.

Embodiments according to the present invention will be described below with reference to the drawings.
<First embodiment>
An anomaly detection system 101 and a detection device 100 incorporated in the anomaly detection system 101 according to the present embodiment will be described with reference to FIGS. 1 to 7. FIG. As shown in FIG. 1, an abnormality detection system 101 is used, for example, in a production line 103 for screws 102 to detect an abnormality in the manufactured screw 102 and sort the screws 102 into non-defective products and non-defective products. An abnormality detection system 101 captures an image of a screw 102 in a production line 103 in which non-defective products and defective products coexist, and detects an abnormality based on the captured image. The abnormality detection system 101 distributes the screws 102 for which no abnormality has been detected to the non-defective production line 103 , and distributes the screws 102 for which the abnormality has been detected to the defective production line 103 . This figure shows that the defective screw 102 has an abnormality in which the tool groove (cruciform portion) of the head is crushed. Of course, the anomaly detection system 101 is not limited to the example shown in this figure, and can be used for various detection targets such as a structure itself, parts such as substrates constituting the structure, and products including food.

As shown in FIG. 2, the anomaly detection system 101 includes a detection device 100 according to the present invention, a recording control unit 204 that controls recording of images and videos processed by the detection device 100, and the recording control unit 204 instructs recording. a recording device 205 for storing the recorded image; a display control unit 206 for controlling display of an image or the like instructed to be displayed by the recording control unit 204; A display output device 207 that displays and outputs images and the like, and a device control device 208 that controls devices such as the production line 103 according to the sorting control performed by the detection device 100 (switches the flow direction, etc.).

The recording control unit 204 uses the results determined by the detection device 100 to control the recording of images and videos, as well as the compression rate and recording interval of recorded videos. The recording device 205 records and retains images and the like obtained from the detection device 100 according to commands from the recording control unit 204 . The display control unit 206 controls display of an image or the like acquired by the detection device 100 , a result determined by the detection device 100 , and information saved in the recording device 205 . The display output device 207 actually displays these images, results, information, and the like.

The detection device 100 is a device that detects normality/abnormality of a detection target included in an image. As shown in the figure, the detection device 100 includes an imaging device 201 that captures an image of a detection target, an image acquisition unit 202 that acquires an image or video (first image) including the detection target captured by the imaging device 201, an image An image processing unit 203 is provided for processing an image based on the image or the like acquired by the acquisition unit 202 and determining whether the image is good or bad. Imaging device 201 is, for example, one or more industrial cameras. As a result, it is possible to provide the detection device 100 integrated with the imaging device 201 , and to quickly detect normality/abnormality from an image or the like obtained by the imaging device 201 . Note that the imaging device 201, the image acquisition unit 202, and the image processing unit 203 do not necessarily have to be integrated. may be located remotely (eg, in a control room) and communicatively connected to each other.

The image acquisition unit 202 acquires the signal obtained from the imaging device 201 as an image or the like. The image acquisition unit 202 obtains a one-dimensional, two-dimensional, or three-dimensional image from video signals input from real-time image data from a camera, which is the imaging device 201, or from a video recording device in which image data is recorded. Get it as data. In this image data, processing such as a smoothing filter may be appropriately performed as preprocessing in order to reduce the influence of flicker and the like. Further, data formats such as RGB color, YUV, and monochrome may be selected according to the application. Furthermore, in order to reduce the processing cost, the image data may be reduced to a predetermined size. Note that the image processing unit 203 will be described later.

Here, an example of the hardware configuration of the anomaly detection system 101 will be described with reference to FIG. The anomaly detection system 101 includes a processing unit Prc composed of processing units such as CPU (Central Processing Unit) and MPU (Micro Processing Unit), memory devices such as ROM (Read Only Memory) and RAM (Random Access Memory), It includes a storage unit Mem such as a storage device such as a hard disk HD or DVD, and a communication unit Com of a network interface for communicating with the imaging device 201 and the device control device 208 to input and output signals. They are connected to each other via a transmission line such as a system bus including an expansion bus.

The processing unit Prc has one or more processors (or cores) and their peripheral circuits capable of executing multiple programs in parallel. The processing unit Prc includes an overall control unit 209 that controls the overall operation of the abnormality detection system 101, transmits and receives control signals and information signals (data) to and from the other components described above, and It performs various kinds of arithmetic processing required for processing, execution, and control of the detection system 101 . Therefore, the processing unit Prc performs arithmetic operations such as addition, subtraction, multiplication, and division using a numerical operation unit or the like, logical operations such as logical product, and vector operations according to the learned model in a storage area that can be accessed at high speed. is configured to allow

The storage unit Mem includes various types of memory devices and storage devices according to usage, and partly configures the recording device 205 . For example, the ROM generally records an IPL (Initial Program Loader) that is executed first after the power is turned on. By reading this into the processing unit Prc and executing it, the programs, data, learned models, etc. stored in the storage device such as the hard disk HD are temporarily stored in the RAM for temporary storage by the overall control unit 209. These programs are written out and executed by the overall control unit 209 . The programs stored in the storage unit Mem are an operating system program, programs and modules necessary for the anomaly detection system 101, and trained models. Note that the operating system program is MICROSOFT (registered trademark) WINDOWS (registered trademark), LINUX (registered trademark), UNIX (registered trademark), or the like, and is not particularly limited as long as the anomaly detection system 101 can be executed.

The processing unit Prc reads programs and the like necessary for the anomaly detection system 101 from the storage unit Mem, and controls the image acquisition unit 202, the image processing unit 203, the recording control unit 204, the display control unit 206 that controls the display output device 207, and the like. It implements the functions of the anomaly detection system 101 . In this way, the above-described hardware and software necessary for the anomaly detection system 101 work together to construct unique processing and operations of the anomaly detection system 101 . The anomaly detection system 101 is not limited to the hardware described above, and may be replaced by a non-computer system such as a Digital Signal Processor (DSP), a Field-Programmable Gate Array (FPGA), or a Graphics Processing Unit (GPU). may

The image processing unit 203 will be described with reference to FIG. The image processing unit 203 receives an image or the like (first image) from the image acquisition unit 202, outputs a determination result to the device control device 208, and outputs an image or the like to the recording control unit 204 using the determination result. . The image processing unit 203 includes an image processing unit 301 that outputs an image (second image) obtained by processing the image (first image) input from the image acquisition unit 202, and an image (third image) processed by a learned autoencoder. AE processing unit 303 that outputs an image), and a calculation that outputs an image (fourth image) obtained by calculating the difference between the input image (first image) and the image processed by the AE processing unit 303 (third image). It includes a processing unit 304 and a determination unit 305 that determines normality/abnormality based on the image (fourth image) for which the difference is calculated.

The image processing unit 301 has a function of variously processing the input image. For example, the image processing unit 301 may process the entire image or may process the image partially. Also, the image processing unit 301 may process the image by various processing methods such as brightness conversion processing, blurring processing, edge enhancement processing, and alpha blending processing. Brightness conversion processing is processing for changing brightness, which is one of the three elements that constitute the color space of an image. Further, the lightness conversion processing may be more detailed processing such as changing saturation for each hue. The blurring process is a process of calculating pixel values (RGB values) of pixels in an image by combining them with surrounding pixels using various filters. Various filters are, for example, a Gaussian filter, an averaging filter, a median filter, and the like. The direction and size of the periphery are appropriately determined.

Edge enhancement processing is processing that converts pixels using various filters so that the image becomes clearer, and is processing that enlarges portions of the image where there is a large change (gradient) in pixel values (luminance). Various filters are, for example, a Prewitt filter, a Sobel filter, a Laplacian filter, a Sharp filter, and the like. The direction for edge detection and the magnitude of the gradient are appropriately determined. The alpha blending process is a process of multiplying an image by an alpha value (transparency information) to superimpose a translucent image. The blended image can be any texture.

The image processing unit 301 may apply the processing method described above to the entire image or to a part of the image. When processing a part of an image, it may be a pattern such as regular vertical stripes, horizontal stripes, or lattice, or may be random dots, islands, or the like. It may have a spiral shape or the like, and is not particularly limited. When the image is partially processed, as will be described later, the detection target (such as a flaw or defect) that was present in the determination target appears in the third image, and the detection target portion is emphasized and displayed in the fourth image. , it is possible to prevent such a detection target from being overlooked.

The image processing unit 301 may output a plurality of images (second images) by processing the input image using a plurality of different processing methods. In this case, the AE processing unit 303 in the next step processes the processed multiple images and outputs multiple images (third images) respectively corresponding to the multiple images. In addition to the processed image, the image processing unit 301 may directly output the input image to the AE processing unit 303 without processing the input image. In this case, the AE processing unit 303 processes at least the processed image (second image) out of the unprocessed image (first image) and the processed image (second image).

When processing an input image using a plurality of processing methods, the image processing unit 301 selects any two or more from a combination of brightness conversion processing, blurring processing, edge enhancement processing, and alpha blending processing. It is preferable to perform different processing treatments. In this way, by performing a plurality of different processings on the input image (first image) to be determined, a plurality of differently processed images can be input to the AE processing unit 303 .

Also, when the image processing unit 301 performs two or more different processing, it is preferable that the processing be performed on different portions of the image. By performing different processing on different portions of the input image (first image) in this way, a plurality of images having different portions processed differently can be input to the AE processing unit 303 . For example, as shown in FIG. 4, the image processing unit 301 performs alpha blending processing with a spiral mask in processing processing 1 on an input image 401, and blurs it in horizontal stripes in processing processing N-1. These are output, and an input image that is not processed is also output. In this manner, the image processing unit 301 performs processing such as intentionally adding noise to the input image, and outputs the processed image 402 .

The AE processing unit 303 processes the processed image 402 output by the image processing unit 301 using a learned autoencoder. An autoencoder is composed of an encoder and a decoder composed of a neural network. The encoder outputs dimensionally compressed features from the input data, and the decoder functions to recover the input data from the features. In the machine learning stage, the input to the autoencoder and the output from the autoencoder are compared to calculate the error, and the error is minimized by, for example, error back propagation, i.e., the output of the autoencoder and the input are matched. Adjust the neural network weights to .

When the autoencoder is used to detect anomalies, etc., as shown in FIG. 8, the autoencoder is trained using only image data of normal products with no anomalies. When the learned autoencoder receives an image of a normal screw, it restores the normal screw and outputs it. Even if an image showing a screw is input, it will try to output an image from which the abnormal part is removed, that is, an image of a normal screw. Therefore, by comparing an image containing an abnormality with an output image that appears to be normal and extracting the difference, it is possible to detect an abnormality by specifying the location of the abnormality. However, if the abnormal portion to be detected in the image is shown in an extremely small area, it is difficult to detect it with a normal autoencoder. The present invention solves such problems.

It should be noted that the structure of the intermediate layer in the learning model 302 of the trained autoencoder is not limited to the depth of layers, the size and number of filters, and the like. Also, the autoencoder may use another technique based on the autoencoder, such as similar DAE or VAE. Also, the learning model 302 of the learned autoencoder is stored in the recording device 205 or the like, and is loaded into the processing unit Prc as part of the image processing unit 203 during operation of the detection device 100 .

As shown in FIG. 4, the AE processing unit 303 receives a processed image 402 that has undergone processing processing 1 by the above-described learned autoencoder, and converts the AE processing result image 403 of the AE processing result 2 to Output. The AE processing result image 403 of the AE processing result 2 is closer to a normal product than the processed image 402 subjected to the processing processing 1, but cannot be completely restored due to the spiral mask, and is partially restored with a sense of incongruity. It is an image. Similarly, the AE processing unit 303 outputs an AE processing result image 403 of the AE processing result N when the processed image 402 subjected to the processing processing N−1 is input. The AE processing result image 403 of the AE processing result N is closer to a normal product than the processed image 402 subjected to the processing processing N−1, but it cannot be completely restored due to the horizontal striped mask, and it is partially uncomfortable. It is a restored image.

Also, in the AE processing result image 403, which is the restored image, an abnormal portion not included in a normal screw (in this figure, the tool groove of the screw head is The (cruciform part) is not completely restored due to the crushed part), and it is partially restored in a strange shape. Of course, the AE processing unit 303 cannot completely restore a part of an image that is not processed, even if the abnormal part is extremely small, and the part of the abnormal part is restored in a strange manner. it is conceivable that. Conversely, it can be said that the AE processing unit 303 restores part of the abnormal portion to a normal, comfortable shape.

When the image processing unit 301 also outputs an input image that is not processed, the AE processing unit 303 processes at least the processed image (third image) out of the unprocessed input image and the processed image. do. Also, as shown in FIG. 3, it is preferable that a plurality of AE processing units 303 be provided in order to parallelize and speed up calculation processing.

As shown in FIG. 5, the calculation processing unit 304 combines an input image 401 (first image) acquired by the image acquisition unit 202 and an AE processing result image 403 (third image) restored and output by the AE processing unit 303. is calculated as a difference result image 404 (fourth image). The difference result image 404 is obtained, for example, by measuring the color space distance for both corresponding pixels and highlighting those pixels where the distance is non-zero or above a predetermined threshold. The predetermined threshold value is appropriately determined so as to make the difference stand out.

In this figure, the calculation processing unit 304 calculates the difference result image 404 of the difference result 2 by calculating the difference between the input image 401 and the AE process result image 403 of the AE process result 2. A difference result image 404 of the difference result 2 is an image showing the difference between the input image 401 and the restored image which cannot be completely restored by the spiral mask and which is partially unnatural. Similarly, the difference result image 404 of the difference result N is an image showing the difference between the input image 401 and the restored image that cannot be completely restored by the horizontal striped mask and is partially unnatural.

Further, the difference result image 404 of the difference result 1, which shows the case where the AE processing unit 303 also processes an image that is not processed, is the same as the input image 401, and even if the abnormal portion is extremely small, a part of it is FIG. 10 is an image showing the difference between a portion that cannot be completely restored and is partially restored in a strange form, and a portion that is different from the abnormal portion and is normally restored in a form that does not give a strange feeling. . In addition, if the input image 401 includes an abnormal portion, the difference included in the difference result image 404 of the difference result 1 may be changed to other difference result images such as the difference result image 404 of the difference result 2 and the difference result image 404 of the difference result N. 404 may also be included. In this figure, the image processing unit 301 outputs a plurality of processed images 402, and the AE processing unit 303 processes the plurality of images and outputs a difference result image 404 which is a plurality of difference results. However, when processing one image, the calculation processing unit 304 outputs one difference result image 404 (fourth image).

When the AE processing unit 303 processes a plurality of images, the calculation processing unit 304 calculates a plurality of difference results obtained by calculating the differences between the input image 401 and the plurality of AE processing result images 403 restored by the AE processing unit 303. An image 404 is generated, and a single synthesized result image 405 (fourth image) is output by synthesizing the plurality of difference result images 404 . Various methods are conceivable for synthesizing the images, but if the purpose is to emphasize the difference common to the plurality of difference result images 404, the calculation processing unit 304 combines the logic of the calculated plurality of difference result images 404. It is preferable to synthesize by calculating the product and output one synthesized result image 405 (fourth image). As a result, the composite result image 405 becomes an image in which even an extremely small detection target is emphasized and displayed by superimposing the relevant portion.

For example, as shown in this figure, the difference included in the difference result image 404 of the difference result 1, that is, the difference due to the abnormal portion, is the difference result image 404 of the difference result 2, the difference result image 404 of the difference result 3, . and the difference result image 404 of the difference result N, one synthesis result image 405 synthesized by calculating the logical product of all these difference result images 404 is extremely large in each difference result image 404. Even a small difference is calculated as a large difference.

When the AE processing unit 303 processes a plurality of images, the calculation processing unit 304 is executed as shown in the flowchart of FIG. In S100, the calculation processing unit 304 acquires a difference result image 404 that is a plurality of difference results 1 to N output by the AE processing unit 303. FIG. In S102, the calculation processing unit 304 converts pixel values equal to or greater than the threshold A to 1 (white) and pixel values that do not exceed the threshold A to 0 ( black). In this way, the portion above the threshold A is emphasized. In S<b>104 , the calculation processing unit 304 performs logical product (AND processing) on each corresponding pixel in each converted difference result image 404 . In order to further emphasize the difference, the calculation processing unit 304 uses different thresholds B, and sets the logical AND result to 1 (white) if each pixel is equal to or greater than the threshold B, and to logical if the threshold B is not exceeded. The result of the product may be 0 (black).

When the AE processing unit 303 processes one image, the determination unit 305 determines normality/abnormality based on one difference result image 404 (fourth image). In this case, one difference result image 404 other than the difference result 1 is output. Further, when the AE processing unit 303 processes a plurality of images, the determination unit 305 performs normal/abnormal determination based on one combined result image 405 (fourth image). In this way, an input image 401 (first image) to be determined and an AE processing result image 403 (third image) obtained by processing an input image 401 (first image) and a processed image 402 (second image) processed by the autoencoder. By making a determination based on the difference result image (fourth image) showing the difference between the normal/ It is possible to provide the detection device 100 that prevents an abnormality from being determined.

In particular, a plurality of different AE processing result images 403 (third images) obtained by processing a plurality of different processed images 402 (second images) by an autoencoder are combined to form one composite result image (fourth image). image) is generated, and determination is made based on the synthesized result image 405 (fourth image), so that even an extremely small detection target can be calculated and displayed in an emphasized manner by superimposing such a detection target. You can avoid missing it. Further, as described above, by performing a plurality of different processing processes on the input image 401 (first image), which is the determination target, all the detection targets such as scratches that existed in the determination target are the AE processing result images 403 (first image). 3 images), even if the detection target is small, it is calculated and displayed in an emphasized manner by being superimposed on the synthesis result image 405 (fourth image), and such a detection target cannot be overlooked. can be prevented.

The above is also a detection method for detecting the normality/abnormality of the detection target included in the image. This detection method includes the steps of acquiring an input image 401 (first image) in which a detection target such as a screw is shown, outputting a processed image 402 (second image) obtained by processing the input image, a step of outputting an AE processing result image 403 (third image) obtained by processing at least the processed image 402 out of the input image 401 and the processed image 402 by the autoencoder; 403 to output a difference result image 404 (fourth image); and a step of determining normality/abnormality based on the difference result image 404. In this way, by making a determination based on the difference result image 404 showing the difference between the input image 401 to be determined and the processed image 402 obtained by processing the input image 401 and the AE processing result image 403 processed by the autoencoder, Even if an object to be detected is shown in an extremely small area in an image, it is possible to provide a detection method that prevents such a detection object from being overlooked and normality/abnormality to be determined.

According to the present invention, in the manufactured product anomaly detection system 101 using an industrial camera or the like, it is possible to not overlook defects such as extremely small scratches, defects, and processing deviations that are mixed in with manufactured products.

It should be noted that the present invention is not limited to the exemplified embodiments, and can be implemented with a configuration that does not deviate from the contents described in each claim. That is, although the present invention has been particularly illustrated and described primarily with respect to particular embodiments, there may be modifications, quantities, Various modifications can be made to other detailed configurations by those skilled in the art.

100: Detecting device, 101: Anomaly detection system, 102: Screw (detection target), 103: Manufacturing line, 201: Imaging device, 202: Image acquiring unit, 203: Image processing unit, 204: Recording control unit, 205: Recording Device 206: Display control unit 207: Display output device 208: Equipment control device 209: Overall control unit 301: Image processing unit 302: Learning model 303: AE processing unit 304: Calculation processing unit 305 : determination unit, 401: input image (first image, determination target), 402: processed image (second image), 403: AE processing result image (third image), 404: difference result image (fourth image ), 405: Synthesis result image (fourth image)

Claims

A detection device for detecting normality/abnormality of a detection target included in an image,
an image acquisition unit that acquires a first image including a detection target;
an image processing unit that outputs a second image obtained by processing the first image;
an AE processing unit that outputs a third image obtained by processing at least the second image out of the first image and the second image by a learned autoencoder;
a calculation processing unit that calculates the difference between the first image and the third image and outputs a fourth image;
a determination unit that determines normality/abnormality based on the fourth image;
A detection device comprising:
The detection device according to claim 1, wherein the image processing unit outputs the second image obtained by partially processing the first image.
The image processing unit outputs a plurality of second images by performing a plurality of different processing processes on the first image,
the AE processing unit processes the plurality of second images and outputs a plurality of third images corresponding to each of the plurality of second images;
The calculation processing unit calculates a difference between the first image and the plurality of third images, and outputs one of the fourth images.
3. The detection device according to claim 1 or 2, characterized in that:
The detection device according to claim 3, wherein the calculation processing unit outputs the first fourth image by calculating a logical product of a plurality of calculated differences.
5. The detection device according to claim 3, wherein the image processing unit performs any two or more different processing from a set of brightness conversion processing, blurring processing, edge enhancement processing, and alpha blending processing. .
further comprising an imaging device for imaging the detection target,
The detection device according to any one of claims 1 to 5, wherein the image acquisition unit acquires the first image from the imaging device.
A detection method for detecting normality/abnormality of a detection target included in an image,
obtaining a first image containing the detection target;
outputting a second image obtained by processing the first image;
outputting a third image obtained by processing at least the second image out of the first image and the second image by a trained autoencoder;
calculating a difference between the first image and the third image and outputting a fourth image;
a step of determining normality/abnormality based on the fourth image;
A detection method comprising: